Toward semi-automatic biologically effective dose treatment plan optimisation for Gamma Knife radiosurgery

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Dose-rate effects in Gamma Knife radiosurgery (GKRS) cause variable biologically effective dose (BED) for identical physical doses.
• The non-convex BED model's dependence on delivery sequence complicates GKRS treatment planning.
• Existing GKRS planning methods may not account for dose-rate variations, impacting treatment consistency.

Purpose of the Study:

• To investigate the feasibility of inverse planning for GKRS to achieve desired BED characteristics.
• To optimize per-isocenter beam-on times and delivery sequence for consistent BED.
• To evaluate computational approaches for generating clinically relevant GKRS treatment plans.

Main Methods:

• Implemented two optimization algorithms: Mixed-Integer Linear Programming (MILP) and a faster local optimization approach.
• MILP utilized a convex underestimator for BED to address local minima, increasing computational complexity.
• Local optimization sequentially optimized beam-on times (quasi-Newton) and delivery sequence (local search).

Main Results:

• Local optimization converged significantly faster (minutes vs. hours/days) than MILP, achieving comparable objective function values (within 1.2%).
• Treatment plan quality metrics showed no meaningful differences between MILP and local optimization.
• Optimized plans eliminated treatment time dependence and promoted more conformal dose distributions.

Conclusions:

• Inverse planning is feasible within a reasonable timeframe for achieving BED-based objectives in GKRS.
• Optimized beam-on times and delivery sequences ensure consistent BED across varying treatment durations.
• This approach holds promise for enhancing GKRS treatment plan quality and consistency.